Purdue UniversityPurdue e-PubsDepartment of Electrical and ComputerEngineering Technical Reports

Department of Electrical and ComputerEngineering

11-6-2017

28-GHz Channel Measurements and Modeling forSuburban EnvironmentsYaguang ZhangPurdue University, ygzhang@purdue.edu

David J. LovePurdue University, djlove@purdue.edu

Nicolo MichelusiPurdue University, michelus@purdue.edu

James V. KrogmeierPurdue University, jvk@purdue.edu

Christopher R. AndersonUnited States Naval Academy, canderso@usna.edu

See next page for additional authors

Follow this and additional works at: http://docs.lib.purdue.edu/ecetr

This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact epubs@purdue.edu foradditional information.

Zhang, Yaguang; Love, David J.; Michelusi, Nicolo; Krogmeier, James V.; Anderson, Christopher R.; Jyoti, Soumya; and Sprintson,Alex, "28-GHz Channel Measurements and Modeling for Suburban Environments" (2017). Department of Electrical and ComputerEngineering Technical Reports. Paper 483.http://docs.lib.purdue.edu/ecetr/483

AuthorsYaguang Zhang, David J. Love, Nicolo Michelusi, James V. Krogmeier, Christopher R. Anderson, SoumyaJyoti, and Alex Sprintson

This article is available at Purdue e-Pubs: http://docs.lib.purdue.edu/ecetr/483

28-GHz Channel Measurements and Modelingfor Suburban EnvironmentsYaguang Zhang, Soumya Jyoti, Christopher R. Anderson,

David J. Love, Nicolo Michelusi, Alex Sprintson, and James V. Krogmeier

Abstract—This paper presents millimeter wave propagationmeasurements at 28 GHz for a typical suburban environmentusing a 400-megachip-per-second custom-designed broadbandsliding correlator channel sounder and highly directional 22-dBi(15◦ half-power beamwidth) horn antennas. With a 23-dBmtransmitter installed at a height of 27 m to emulate a microcelldeployment, the receiver obtained more than 5000 power delayprofiles over distances from 80 m to 1000 m at 50 individuals sitesand on two pedestrian paths. The resulting basic transmissionlosses were compared with predictions of the over-rooftop modelin recommendation ITU-R P.1411-9. Our analysis reveals that thetraditional channel modeling approach may be insufficient to dealwith the varying site-specific propagations of millimeter waves insuburban environments. For line-of-sight measurements, the pathloss exponents obtained for the close-in (CI) free space referencedistance model and the alpha-beta-gamma (ABG) model are 2.00and 2.81, respectively, which are close to the recommended site-general value of 2.29. The root mean square errors (RMSEs) forthese two reference models are 9.93 dB and 9.70 dB, respectively,which are slightly lower than that for the ITU site-general model(10.34 dB). For non-line-of-sight measurements, both referencemodels, with the resulting path loss exponents of 2.50 for the CImodel and 1.12 for the ABG model, outperformed the site-specificITU model by around 14 dB RMSE.

Index Terms—Millimeter wave, channel measurements, chan-nel modeling, suburban environments

I. INTRODUCTION

The increasing demand of mobile device users for higherdata rates has been the driving factor for the rapid developmentof mobile telecommunications during the past decade [1]. Thenumber of worldwide mobile subscriptions, which reached arecord height of 7.5 billion in 2016, has continued to increase,and the total mobile data traffic is expected to rise at acompound annual growth rate of 42% between the end of 2016and 2022 [2]. This increasing usage of wireless technologyis prompting mobile service providers to take advantage ofhigher frequency bands in the foreseeable future, startingwith millimeter waves (mm-waves), to overcome the expectedglobal bandwidth shortage [3].

Y. Zhang, D. J. Love, N. Michelusi, and J. V. Krogmeier are with the Schoolof Electrical and Computer Engineering, Purdue University, 465 NorthwesternAvenue, West Lafayette, IN 47907, USA. (Email: {ygzhang, djlove, michelus,jvk}@purdue.edu)

C. R. Anderson is with the Department of Electrical and ComputerEngineering, United States Naval Academy, 105 Maryland Ave, Annapolis,MD 21402, USA. (Email: canderso@usna.edu)

S. Jyoti and A. Sprintson are with the Department of Electrical andComputer Engineering, Texas A&M University, College Station, TX 77843,USA. (Email: {soumyajyoti.jh, spalex}@tamu.edu)

Sponsorship for this work was provided by NSF research grant 1642982.

With recent advances in radio frequency (RF) technol-ogy [4]–[6], hardware operating at mm-wave frequencies isbecoming commercially available [7]–[9]. This has made mm-wave frequencies the most promising higher frequency bandsfor a larger usable radio spectrum. To fully understand thepropagation characteristics of mm-wave signals, the amountof research on mm-wave channel measurement and modelinghas increased dramatically in the past five years [1], [10]–[12]. However, the majority of current research on mm-wave communications has focused on urban areas with highpopulation densities, with very few measurement campaignsfor suburban and rural environments, which are also importantfor future mobile networking technologies. Moreover, point-to-point links have received significant attention, but the channelinformation provided by this approach is insufficient for next-generation wireless networks such as 5G. These networks mustaddress mobility requirements because frequent and location-specific blockages are expected for mm-waves, which requirea much richer set of channel state information beyond LoSpath loss (such as multipath scattering).

In this paper, we explore this gap in the research by fo-cusing on suburban environments and constructing physically-motivated and practical models that can be used for mm-wavenetworks in these environments. An intensive measurementcampaign has been carried out at the United States NavalAcademy (USNA) in Annapolis, Maryland. Measurementswere taken around the campus at 28 GHz to characterize thepropagation in a suburban-type environment. The resultingpath losses are compared with several standard 5G channelmodels. Our results indicate that a holistic, network-wide ap-proach beyond traditional point-to-point links may be neededto deal with the high dependence of mm-waves on site-specificenvironment geometry.

The paper is organized as follows: In Section II, we presentour propagation measurements in suburban environments; InSection III, we present the procedure for calculating basictransmission losses, followed by our analysis in Section IV;Finally, in Section V, we conclude the paper.

II. MM-WAVE PROPAGATION MEASUREMENTSFOR SUBURBAN ENVIRONMENTS

An outdoor propagation measurement campaign was con-ducted on the USNA campus to emulate a typical 5G de-ployment in a suburban environment. The transmitter (TX)was temporarily installed on the Mahan Hall clock tower. The

PNSequenceGenerator

WindFreakSynthNV

400 MHz 0 dBm

WindFreakSynthNV

2.5 GHz +7 dBm

28-GHz

Upconverter

MiniCircuitsSBLP-933+

DC–900 MHz

10 dB MiniCircuitsZFL-1000+

10–1000 MHz17 dB Gain

3 dB 3 dB MiniCircuitsVBF-2435+

2340–2530 MHz

WR-28Horn Antenna+22 dBi Gain

LO

IF RF

(a) Transmitter

WR-28Horn Antenna+22 dBi Gain

28-GHz

Downconverter

MiniCircuitsVBF-2435+

2340–2530 MHz

IF

RF LO

MiniCircuitsSHP-900

>900 MHz

MiniCircuitsVBF-2435+2340–2530 MHz

USRP

B200

Laptop for Storage

PNSequenceGenerator

WindFreakSynthNV

399.94 MHz 0 dBm

MiniCircuitsSBLP-933+

DC–900 MHz

9 dB MiniCircuitsZFL-1000+

10–1000 MHz

3 dB

17 dBGain

(b) Receiver

Fig. 1. Block diagrams for the 28-GHz broadband sliding correlator channel sounder. Model numbers are labeled for some commercially available parts.

receiver (RX) was moved around the campus to obtain pathloss measurements for individual sites and continuous paths.

A. Measurement Equipment

A custom-designed broadband sliding correlator channelsounder [13] was used to record propagation data. Figure 1presents block diagrams for the channel sounder. At both theTX and the RX sides, horn antennas with a nominal +22 dBigain and 15◦ half-power beamwidth (HPBW) were employed,and pseudo-random noise (PN) generators produced the samePN sequences with a chip sequence length of 2047 [1].

At the TX, the PN probing signal was generated with achip rate of 400 megachips per second (Mcps). It was firstmodulated to a 2.5-GHz intermediate frequency (IF) and thenconverted to RF of 28 GHz by mixing it with a 25.5-GHz localoscillator (LO) at the upconverter. At the RX, the receivedsignal was first downconverted from 28 GHz to 2.5 GHz andthen cross-correlated with the identical PN sequence generatedwith a slightly slower clock rate of 399.94 MHz, similarto the setup in [1]. A Universal Software Radio Peripheral(USRP) B200 was utilized to record the resulting in-phase(I) and quadrature (Q) signal components. It also regularlysampled the RX’s location with the help of an on-board GPSdisciplined oscillator (TCXO version). The implementation ofthe measurement system is shown in Figure 2.

Table I summarizes the key parameters for the channelsounder. Note that the estimation for the maximum measurablepath loss is based on the following. (1) The RX low-noiseamplifier has a noise figure of 2.4 dB, so we assume a worst-case RX noise figure of 6 dB. (2) The RX detection bandwidthis 60 kHz. (3) The estimated minimum signal-to-noise ratio(SNR) for a detectable signal is 5 dB.

(a) TX (b) RX

Fig. 2. Photos of the custom-designed spread-spectrum channel sounder.Components are fixed on shelves for portability and the TX is further moreenclosed into a rack case for extra protection during the deployment.

TABLE IBROADBAND SLIDING CORRELATOR CHANNEL SOUNDER

SPECIFICATIONS

Carrier Frequency 28 GHz

Chip Sequence Length 2047

RF Bandwidth (First Null) 800 MHz

TX Chip Rate 400 Mcps

Temporal Resolution 2.5 ns

RX Chip Rate 399.94 Mcps

TX Power 23 dBm

TX/RX Antenna Gain 22 dBi

Measured TX/RX Azimuth HPBW 10.1◦

Measured TX/RX Elevation HPBW 11.5◦

Maximum Measurable Path Loss 182 dB

B. Measurement Setup and Procedure

Three types of measurements have been performed forlarge-scale path loss, emulated single-input and multiple-

(a) RX on an Electric Car

(b) RX on a Trolley

(c) Before a Continuous SignalRecording for a Pedestrian Path

Fig. 3. Photos of the measurement setup. The TX was installed on a clocktower at a height of 90 feet (27.4 m) to emulate a microcell deployment.Either an electric car or a two-layer trolley was used to transport the RX. Forboth cases, the platform for the RX could be rotated horizontally and tiltedwith composite wood shims to align the beam.

output (SIMO), and continuous tracks. The TX was installed ata height of 90 feet (27.4 m) to emulate a microcell deployment.The RX was moved around campus by an electric car or atwo-layer platform trolley to obtain the measurements, whichis illustrated by the photographs in Figure 3.

We used compasses and digital levels to achieve beamalignment prior to the measurements at each RX location.Aside from the USRP output files and GPS samples, theazimuths and elevations of both the TX and RX antennaswere recorded manually. This allowed us to reconstruct thegeometry relationship of the antennas and extract preciseantenna gains. The USRP gain was manually adjusted tomaximize the SNR at the RX.1

1) Large-Scale Path Loss Measurements: Forty measure-ment locations were chosen to investigate the large-scale pathloss at various distances from the TX (between 100 m and1000 m). For each site, the RX antenna was moved along Xand Z axes by our positioning system to form a “+” patternwithin a 20λ × 20λ area, where λ = 10.7 mm was the wavelength corresponding to 28 GHz. For every λ interval onthe “+” pattern, one 3-second signal recording and one GPSsampling were obtained, with 40 separate measurements intotal. In Figures 3b and 3c, the custom-built X-Z positioningsystem is shown.

1The Python code for automatically carrying out the measure-ments at each location using GNU Radio, together with the MAT-LAB code for post-processing the collected data, are available athttps://github.com/YaguangZhang/EarsMeasurementCampaignCode.git

2) Emulated SIMO Measurements: Measurements thatwere used to emulate SIMO signals were similar to thosefor the large-scale path loss, but with a higher space sampledensity and a larger sample area for each site. The RX antennawas moved along the “+” pattern with the interval betweenadjacent measurements reduced to 0.25λ. The pattern coveredan enlarged 40λ× 40λ area, with 320 separate measurementsin total for each site. Ten locations were chosen, and theTX and RX distances varied from 50 m to 500 m. Thedata collected were or will be used to analyze small-scalepropagation characteristics. However, in this paper, we focusonly on the large-scale path loss.

3) Measurements for Continuous Tracks: To investigatethe shadowing effect on a moving user, two approximately200-m-long straight tracks were chosen for continuous signalrecordings. The RX antenna was fixed on the positioningsystem and moved at walking speed along each track to recordthe signal, and the GPS location of the RX was sampled onceper second. The RX antenna elevation was kept at 0◦, and theplatform was adjusted as necessary to maintain the azimuthwith respect to the fixed earth reference during the recordingprocess (see Figure 3c).

III. BASIC TRANSMISSION LOSS CALCULATION

In this section, we present our procedure for computingthe basic transmission loss of each signal recording. Receivedsignal power calculation, RX calibration, antenna patterngeneration, and antenna gain extraction were considered indetermining antenna-independent path losses.

A. Received Signal Power Calculation

The traditional method of calculating received signal powerfor broadband measurements is carried out in the time domainby integrating the total area under a power delay profile(PDP), assuming that all the multipath signals add up inphase. Instead, we adopted a frequency-domain technique, theprocess of which is summarized below, to preserve the phaserelationships and yield a more accurate estimate of receivedpower.

1) Run the recorded complex I/Q voltage waveformthrough a 60 kHz low pass filter (LPF) to remove noisein frequency.

2) Further eliminate noise in the time domain by (a)estimating the standard deviation σ and mean n0 forthe noise using 10%–20% of the samples between twoadjacent signal pulses, as illustrated in the top plot ofFigure 4a; (b) calculating the noise elimination thresholdas 3.5σ + n0; and (c) Setting all samples with anamplitude below the threshold to 0, as illustrated by theplot at the bottom of Figure 4a.

3) Calculate the received power by integrating over thepower spectrum density (PSD) of the noise-eliminatedsignal. We chose a tighter integration range (1 Hz–20 kHz) to ignore residual high-frequency noise anddirect current (DC) components, as illustrated in Fig-ure 4b.

(a) Noise Variance Estimation and Noise Elimination

(b) Integral in Frequency for Signal Power

Fig. 4. Received signal power calculation. (a) For each measurement, thenoise amplitude characteristics were first estimated for the filtered signal toset a noise elimination threshold. Then, all samples weaker than the thresholdwere set to 0. (b) The received power energy was computed by integratingthe PSD of the noise-eliminated signal below 20 kHz, ignoring the DCcomponent.

The resulting values would have a linear relationship withthe actual received power. Next, we needed the RX calibrationto convert the calculated power to the received power.

B. RX Calibration

To calibrate the RX, the upconverter and downconverter,as well as the antennas, were removed from our system inFigure 1. Then, an adjustable attenuator for IF signals with a5-dB step size was attached to the TX to simulate the channel.The RX together with a spectrum analyzer were used to recordand measure the signal strength separately for a set of differentattenuation values, with the results from the spectrum analyzertreated as the reference for signal strengths. We repeated theprocedure with the RX gain (i.e., the adjustable USRP gain)set as its minimum (1 dB) and maximum (76 dB) values.From the obtained power value pairs, linear relationships wereextracted for expected measured power vs. calculated powerat the RX gain limits [14]. Figure 5 shows the resultingtwo base reference calibration lines obtained via orthogonalleast squares, as well as all the calibration lines needed forour whole dataset gotten via a linear interpolation over the

-80 -60 -40 -20

Calculated Power (dB)

-70

-60

-50

-40

-30

-20

Mea

sure

d P

ower

(dB

)

y = 0

.996

77x-

2.90

72

y = 1

.026

9x+5

.797

8

1 dB RX gain76 dB RX gain

(a) Base Calibration Lines

-70 -60 -50 -40 -30 -20

Calculated Power (dB)

-70

-60

-50

-40

-30

Mea

sure

d P

ower

(dB

)

Calibration PtsRef Cali LinesCali Lines Needed

(b) Linearly Interpolated Calibration Lines

Fig. 5. Linear fitting and interpolation for RX calibration lines. (a) Bycarrying out orthogonal least squares over the calibration data points, a linearrelationship was constructed between the calculated power and the referencepower measured by the spectrum analyzer. (b) Calibration lines needed forthe whole dataset were obtained via a linear interpolation over the RX gainwith fixed unit slopes.

RX gain. With these results, received powers were computedaccordingly.

C. Antenna Pattern Generation

Using calibrated measurements from the manufactureras a reference, we were able to account for upcon-verter/downconverter and antenna gains.

1) Normalization for Antenna Plane Patterns: To constructthe pattern for our antennas, two measurement sweeps wereconducted in an anechoic chamber at USNA for the azimuthand elevation planes, respectively, using an Anritsu VectorStarMS4640 series Vector Network Analyzer and the DiamondEngineering DAMS Antenna Measurement System. The for-ward transmission coefficients |S21| from each sweep werenormalized to the nominal maximum gain of the antennas at28 GHz, +22 dBi, and the resulting antenna patterns are plottedin Figure 6.

2) Full 3D Antenna Pattern Filling: From the azimuth andelevation antenna patterns, we constructed a full 3D pattern forthe antenna via linear interpolation, as illustrated in Figure 7.

3) Antenna Gain Computation: With the detailed measure-ment records collected, we were able to reconstruct the geom-

(a) Azimuth Plane Pattern (b) Elevation Plane Pattern

Fig. 6. Antenna pattern measurement results for the azimuth and elevationplanes. The antenna beamwidths (10.1◦ azimuth HPBW and 11.5◦ elevationHPBW) were computed accordingly.

Fig. 7. Full 3D antenna pattern filling.

etry relationship of the TX and RX antennas for each measure-ment and extract their gains. Figure 8 gives an illustration forcomputing the TX antenna gain for one measurement. As wecan see, the Universal Transverse Mercator (UTM) coordinatesystem (x, y) was extended with altitude z to form a three-dimensional (3D) reference system (x, y, z). Inside, a new 3Dcoordinate system (X,Y, Z) was constructed originating at theantenna of interest, which could be for either the TX or theRX, so that the azimuth and elevation for any destination pointcould be calculated. The corresponding gain was extracted byinterpolating the antenna measurement data in 3D.

IV. MEASUREMENT RESULTS AND ANALYSIS

In Figure 9, 49 of the 50 static sites are shown on a Googlemap of the USNA campus. One large-scale site was ignored inour analysis because the data collected there were influencedby rain.

A. Line-of-Sight (LoS) Sites

For LOS sites (24 in total), we compared the measurementresults with the International Telecommunication Union (ITU)site-general model for propagation over rooftops [15]:

PL(d, f) = 10 · α · log10(d) + β + 10 · γ · log10(f)+N(0, σ), (1)

where d is the 3D direct distance between the TX and RX inmeters and f is the operating frequency in GHz. In our case,

f = 28 GHz. The parameters α, β, γ, and σ are chosen forthe LoS propagation in a suburban environment [15]:

α = 2.29

β = 28.6

γ = 1.96

σ = 3.48 .

(2)

These parameters are recommended for a distance range 55 mto 1200 m at 2.2–73 GHz frequency.

For a better comparison, we also considered two othermodels. The first is the close-in (CI) free space referencedistance path loss model:

PL(d) = PLFS(d0) + 10 · n · log10(d

d0

), (3)

PLFS(d0) = 20 · log10(4πd0λ

), (4)

where n is the path loss exponent, PLFS(d0) is the free-spacepropagation loss at a distance of d0 = 1 m with isotropicantennas, and λ is the carrier wavelength. The second isthe alpha-beta-gamma (ABG) model on which the ITU site-general model is based:

PL(d) = 10 · α · log10(d) + β + 10 · γ · log10(f). (5)

Note that α here acts as the path loss exponent.We have fit these two reference models to our LoS mea-

surement results. To deal with GPS sample errors, the medianvalues of latitude, longitude, and altitude measurements foreach site were used as that site’s geographic location tocompute the site distance to the TX. And for the ABG model,we set γ = 1.96 as recommended by ITU, given that we hadonly one carrier frequency at 28 GHz.

All three models, together with the measurement results,are presented in Figure 10a. The ITU model provides pathloss predictions in the form of Gaussian variables, and their 3-sigma range covers the measured path losses reasonably well,even though if we extrapolate the ITU model to d = d0 = 1 m,its mean path loss would be smaller than the correspondingfree-space propagation loss (FSPL). This issue affects theABG type of models in general. In our case, the ABG modelhas the lowest root mean square error (RMSE) at 9.70 dB, butits predicted path loss quickly descends below FSPL as thedistance decreases. Hence, the ABG model does not generalizewell outside the distance range for its measurement data.

The resulting path loss exponents for the CI and ABGreference models are 2.00 and 2.81, respectively. For the LoSdataset, these two models perform 0.41 dB and 0.64 dB betterthan the ITU model, respectively, in terms of RMSE. Basically,within the distance range of the LoS measurement dataset,these three models are equivalent, and the ITU model performswell..

The four sites with lower path loss than the ITU 3-sigmarange were further investigated through their PDPs. As verystrong multipath components showed up for all of thesesites, it is still possible for even narrow mm-waves to have

multipath components that dramatically enhance the signal.For the closer two sites, this was expected because they weresurrounded by buildings. However, the two sites at fartherdistances were in open areas, and the most reasonable originsof the multipath would be reflections from buildings close tothe TX as many of these buildings have sloped ceilings.

B. Non-Line-of-Sight (NLoS) Sites

For the 25 NLoS sites, the site-specific model for prop-agation over rooftops defined in ITU-R P.1411-9, Section4.2.2.2 [15] was utilized to predict the path loss for eachsite. Similar to the LoS case, the CI and ABG models were

compared with the ITU model as references. This time, γ wasset to 2.30 for the ABG model, as recommend by ITU forNLoS above-rooftop propagation from 260 m to 1200 m at2.2–66.5 GHz in an urban high-rise environment, which wasthe most suitable for our purposes.

Figure 10b shows the ITU predictions together with thereference models and NLoS measurement results. For theNLoS data, the ITU model shows a trend of following themeasured path loss, but it has over-estimated predictions formost of the sites. Only five of the NLoS sites had measuredpath losses that were clearly larger than the corresponding

Lon

Lat

Origin

Dest

(a) TX and RX on Map (b) TX Antenna Gain Computation in 3D

Fig. 8. Illustration for antenna gain computation. First, within an extended UTM coordinate system, the azimuth and elevation for the destination, from theorigin’s point of view, were calculated. Then, the resulting gain for the antenna located at the origin were extracted via 3D antenna pattern interpolation. Forthe case illustrated, the antenna of interest was the TX on the clock tower. From its point of view, i.e. in the tilted coordinate system (X,Y, Z) which wasdetermined by the TX antenna’s orientation, the destination has an azimuth of 15.2◦ and an elevation of -3.9◦. The corresponding gain by the 3D interpolatedantenna pattern was 11.05 dB.

Fig. 9. Overview for the basic transmission losses of the large-scale and SIMO measurements. The geographic locations for the measurementsites are shown on a Google map of the USNA campus, with their corresponding worst-case basic transmission losses illustrated by color labels.

TABLE IIKEY PARAMETERS FOR CHANNEL MODELS

ModelLoS NLoS

n α β γ RMSE (dB) n α β γ RMSE (dB)

ITU N/A 2.29 28.6 1.96 10.34 N/A N/A N/A N/A 25.33

Close-in 2.00 N/A N/A N/A 9.93 2.50 N/A N/A N/A 11.73

ABG N/A 2.81 11.66 1.96 9.70 N/A 1.12 63.61 2.30 11.05

1 10 100 200 500 1000

Site Distance to Tx (m)

60

70

80

90

100

110

120

130

140

Path

Lo

ss (

dB

)

Root Mean Square Error:

ITU Mean 10.34 dB

Close-in 9.93 dB

Alpha/Beta/Gamma 9.70 dB

(dB) 79.2

84.5

89.8

95.1

100

106

111

116

121

127

132

ITU MeanITU 3-sigma RangeITU ExtrapolatedClose-inAlpha/Beta/Gamma

(a) For the LoS Meaurements

1 10 100 200 500 1000

Site Distance to Tx (m)

60

70

80

90

100

110

120

130

140

150

160

Path

Lo

ss (

dB

)

Root Mean Square Error:

ITU Mean 25.33 dB

Close-in 11.73 dB

Alpha/Beta/Gamma 11.05 dB

(dB) 91.6

98.4

105

112

119

126

133

139

146

153

160

ITU NLoSClose-inAlpha/Beta/Gamma

(b) For the NLoS Meaurements

Fig. 10. Model comparison for LoS and NLoS measurements. (a) In the LoScase, the ITU predictions agree reasonably well with our measurement results,as well with the reference models. (b) For NLoS sites, the ITU predictionsshow a trend of following the the measurement results but overestimate thepath loss for most of these sites.

ITU predictions. Furthermore, the ABG model does not agreewell with the close-in one, with resulting path loss exponentsof 1.37 and 2.50, respectively. This may be caused by anover-fitting of the ABG model to our NLoS path losses.However, both of these reference models outperformed theITU predictions by around 14 dB RMSE. Table II summarizes

the key parameters for the obtained channel models.Several factors may have reduced the performance of the

ITU site-specific model. First, the buildings on the USNAcampus are not geometrically arranged exactly like thosedefined in the ITU suburban area propagation model. The over-rooftop propagations often span multiple building rows, whichalters the blockage and reflection conditions. Also, for manyof our sites, the ITU over-rooftop model with a limited numberof buildings would be a better fit, which may ameliorate thepath loss overestimation for propagations beyond the buildingarea.

Second, the geometry defined in the ITU model does notalways agree with that for our NLoS sites. For example, thestreet width for most of our sites was over the defined limit.For these cases, we used the upper bound for the model 25 min the calculation, so the model may have underestimated thepath losses for sites relatively close to the TX by consideringreflections from buildings that do not exist. For sites that werefurther away, more buildings are positioned between the TXand the RX, which may have caused an overestimation.

Third, at around 10 of the NLoS sites, the LoS propagationwas directly blocked by vegetation, not buildings. A few othersites have foliage near their LoS paths. ITU-R P.1411-9 doesnot consider the propagation effects caused by vegetation dueto the complexity.

C. Continuous Tracks

Two continuous signal recordings were conducted to in-vestigate the shadowing effect for a simulated moving RX atwalking speed. The basic transmission losses were calculatedfor each second of signal recording (see Figure 11).

For the track on Holloway road (Figure 11b), the mostsignificant blockage was from Michelson Hall, which causedapproximately 30 dB of additional path loss to half of thetrack. Rickover Hall partially blocked the track by around 20dB. The dotted lines in Figure 11b show where these blockagesstarted on the track. There were also metal stadium benches onthe northeast side of Ingram Field adjacent to the measurementroute, which caused around 20 dB of additional path loss.This is more clear on the part of the track with no buildingblockages, where a path loss peak shows up behind each oneof the benches.

For the track on Upshur road (Figure 11c), the two benchesnext to the measurement route were farther away from thetrack and did not block the signal significantly. For this set ofdata, the effect of foliage shadowing is of interest. Trees were

Lon

Lat

92.4

99.4

106

113

120

127

134

141

148

155

162

TX

(dB)

(a) Basic Transmission Losses for Continuous TracksShown on a Google Map of the USNA Campus

Lon

La

t

54.5

60.9

67.3

73.8

80.2

86.6

93

99.4

106

112

119

TXRickover

Hall

Michelson

HallBenches

(dB)

(b) Track on Holloway Road

Lon

La

t

104

108

111

115

119

122

126

130

133

137

141

TXBenches

Trees

(dB)

(c) Track on Upshur Road

Fig. 11. Basic transmission losses for continuous tracks. The path losseson Holloway road illustrate the shadowing effect of buildings and benches,while those on Upshur road illustrate the shadowing effect of trees. In (c),the starting part of the track is zoomed in, and trees are indicated by greenrectangles.

planted on the TX side adjacent to the pathway, and a fewpath loss peaks shown on Figure 11c correspond well withthe locations of these trees.

D. Discussion

We have focused on large-scale path loss, comparing ourmeasurements with predictions obtained by following ITU-RP.1411-9 [15] on propagation over rooftops. Our data showsthat a richer set of channel state information, including multi-path scattering and location-specific building geometries, arenecessary to produce more reliable coverage predictions. Inthe LoS situation, the measurements agree reasonably wellwith the ITU predictions, while for the NLoS situation, mostof the predictions appear to be higher than the corresponding

measurement results. Limiting the building number, extendingthe model with sloped ceilings, and considering over-rooftoppropagations spanning multiple building rows may help im-prove the ITU site-specific model for suburban environments.

According to our model comparison results, the CI andABG models do a good job of predicting large-scale path lossin both LoS and NLoS cases. Furthermore, the ABG model,in general, is able to provide a better fit for a given set of data.However, it may give non-physical results beyond the distancerange over which the data are collected. Because of this, careshould be taken when using the ABG model out of its definedrange.

The continuous recordings back up channel modeling at-tempts that consider environmental information, for example,through the radio environment map [16] and ray tracing [17]approaches. Moreover, the results indicate the possibilityof increasing large-scale path loss prediction accuracy forstatistical models with simple but site-specific environmentgeometry, which would balance computation complexity andmodel performance.

E. Future Works

In the future, we plan to: (1) Further investigate the ITUchannel modeling recommendation for possible extensions orimprovements, e.g. sloped ceilings, over roof-tops propaga-tions spanning multiple rows, and limitation on the buildingnumber; (2) Explore the possibility of enhancing statisticalchannel model performance with side geometric information;(3) Examine the influence on large-scale path loss from thevegetation; (4) Look closer into the SIMO measurements formeans to utilize the presented multipath effects.

V. CONCLUSION

In this paper, we discussed the implementation of a custom-designed broadband channel sounder and explained how weused it for a measurement campaign that focused on the prop-agation of mm-waves in suburban environments at 28 GHz.The resulting basic transmission losses for LoS and NLoS siteswere separately compared with the corresponding propagationpredictions based on the ITU-R P.1411-9 recommendation.The site-general LoS model for propagation over rooftops insuburban environments agreed with our measurements rea-sonably well, but the corresponding site-specific NLoS modeloverestimated the path loss for most of the NLoS sites. Twocontinuous measurement tracks were also constructed. Theresults illustrated that a knowledge of geometric features mayincrease the prediction performance for large-scale path losses,which backs up the radio environment map approach forchannel modeling in future communication networks.

ACKNOWLEDGMENT

We would like to thank the faculty members from theDepartment of Electrical and Computer Engineering at theUnited States Naval Academy for their kind assistance duringthe measurement campaign.

REFERENCES

[1] T. S. Rappaport et al., “Millimeter wave mobile communications for 5gcellular: It will work!” IEEE access, vol. 1, pp. 335–349, 2013.

[2] “Ericsson mobility report june 2017,” Ericsson, SE-164 80 Stockholm, Sweden, Tech. Rep., June 2017.[Online]. Available: https://www.ericsson.com/assets/local/mobility-report/documents/2017/ericsson-mobility-report-june-2017.pdf

[3] T. S. Rappaport, R. W. Heath Jr, R. C. Daniels, and J. N. Murdock,Millimeter Wave Wireless Communications. Pearson Education, 2014.

[4] K. Okada et al., “Full four-channel 6.3-gb/s 60-ghz cmos transceiverwith low-power analog and digital baseband circuitry,” IEEE Journal ofSolid-State Circuits, vol. 48, no. 1, pp. 46–65, 2013.

[5] N. Saito et al., “A fully integrated 60-ghz cmos transceiver chipset basedon wigig/ieee 802.11 ad with built-in self calibration for mobile usage,”IEEE Journal of Solid-State Circuits, vol. 48, no. 12, pp. 3146–3159,2013.

[6] D. Fritsche, G. Tretter, C. Carta, and F. Ellinger, “Millimeter-wavelow-noise amplifier design in 28-nm low-power digital cmos,” IEEETransactions on Microwave Theory and Techniques, vol. 63, no. 6, pp.1910–1922, 2015.

[7] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broad-band systems,” IEEE Communications Magazine, vol. 49, no. 6, 2011.

[8] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-wave cellularwireless networks: Potentials and challenges,” Proceedings of the IEEE,vol. 102, no. 3, pp. 366–385, 2014.

[9] W. Roh et al., “Millimeter-wave beamforming as an enabling technologyfor 5g cellular communications: Theoretical feasibility and prototyperesults,” IEEE Communications Magazine, vol. 52, no. 2, pp. 106–113,2014.

[10] G. R. MacCartney and T. S. Rappaport, “73 ghz millimeter wavepropagation measurements for outdoor urban mobile and backhaulcommunications in new york city,” in IEEE International Conferenceon Communications (ICC), 2014. IEEE, 2014, pp. 4862–4867.

[11] A. I. Sulyman, A. T. Nassar, M. K. Samimi, G. R. Maccartney, T. S.Rappaport, and A. Alsanie, “Radio propagation path loss models for5g cellular networks in the 28 ghz and 38 ghz millimeter-wave bands,”IEEE Communications Magazine, vol. 52, no. 9, pp. 78–86, 2014.

[12] S. Deng, M. K. Samimi, and T. S. Rappaport, “28 ghz and 73 ghzmillimeter-wave indoor propagation measurements and path loss mod-els,” in IEEE International Conference on Communication Workshop(ICCW), 2015. IEEE, 2015, pp. 1244–1250.

[13] T. S. Rappaport et al., Wireless Communications: Principles and Prac-tice. New Jersey: Prentice Hall PTR, 1996, vol. 2.

[14] C. D. Rowe, “Channel propagation model for train to vehicle alertsystem at 5.9 ghz using dedicated short range communication,” Ph.D.dissertation, Blacksburg, VA, 2016.

[15] ITU, “Propagation data and prediction methods for the planning ofshort-range outdoor radiocommunication systems and radio local areanetworks in the frequency range 300mhz and 100ghz,” Tech. Rep.,June 2017. [Online]. Available: https://www.itu.int/rec/R-REC-P.1411-9-201706-I/en

[16] H. B. Yilmaz, T. Tugcu, F. Alagoz, and S. Bayhan, “Radio environmentmap as enabler for practical cognitive radio networks,” IEEE Commu-nications Magazine, vol. 51, no. 12, pp. 162–169, 2013.

[17] T. A. Thomas, H. C. Nguyen, G. R. MacCartney, and T. S. Rappaport,“3d mmwave channel model proposal,” in IEEE 80th Vehicular Tech-nology Conference (VTC Fall), 2014. IEEE, 2014, pp. 1–6.